sign in sign up
<<
Home , Connectogram, Human brain

University of Pennsylvania ScholarlyCommons

Neuroethics Publications Center for & Society

10-2015

Progress and Challenges in Probing the

Russell A. Poldrack

Martha J. Farah University of Pennsylvania, [email protected]

Follow this and additional works at: https://repository.upenn.edu/neuroethics_pubs

Part of the Bioethics and Medical Ethics Commons, Neuroscience and Neurobiology Commons, and the Commons

Recommended Citation Poldrack, R. A., & Farah, M. J. (2015). Progress and Challenges in Probing the . Nature, 526 (7573), 371-379. http://dx.doi.org/10.1038/nature15692

This paper is posted at ScholarlyCommons. https://repository.upenn.edu/neuroethics_pubs/136 For more information, please contact [email protected]. Progress and Challenges in Probing the Human Brain

Abstract Perhaps one of the greatest scientific challenges is ot understand the human brain. Here we review current methods in human neuroscience, highlighting the ways that they have been used to study the neural bases of the human . We begin with a consideration of different levels of description relevant to human neuroscience, from molecules to large-scale networks, and then review the methods that probe these levels and the ability of these methods to test hypotheses about causal mechanisms. Functional MRI is considered in particular detail, as it has been responsible for much of the recent growth of human neuroscience research. We briefly er view its inferential strengths and weaknesses and present examples of new analytic approaches that allow inferences beyond simple localization of psychological processes. Finally, we review the prospects for real-world applications and new scientific challenges for human neuroscience.

Keywords , psychiatric disorders

Disciplines Bioethics and Medical Ethics | Neuroscience and Neurobiology | Neurosciences

This working paper is available at ScholarlyCommons: https://repository.upenn.edu/neuroethics_pubs/136 REVIEW doi:10.1038/nature15692

; Progress and challenges in probing < the human brain = Russell A. Poldrack1 & Martha J. Farah2

> Perhaps one of the greatest scientific challenges is to understand the human brain. Here we review current methods in human neuroscience, highlighting the ways that they have been used to study the neural bases of the human mind. We ? begin with a consideration of different levels of description relevant to human neuroscience, from molecules to large-scale networks, and then review the methods that probe these levels and the ability of these methods to test hypotheses about causal mechanisms. Functional MRI is considered in particular detail, as it has been responsible for much of the recent growth of human neuroscience research. We briefly review its inferential strengths and weaknesses and present examples of new analytic approaches that allow inferences beyond simple localization of psychological processes. Finally, we review the prospects for real-world applications and new scientific challenges for human neuroscience.

he way that we conceptualize brain function has always been of one centimetre. The second characteristic is the ability of each method constrained by the methods available to study it. Studies of to elucidate the mechanistic role of an observed brain molecule, , T patients with focal brain lesions in the nineteenth century led region or network in a mental function of interest. By mechanism we to the view of the brain as a collection of focal centres specialized for mean the causal chain of events that result in the realization of a func- particular cognitive abilties, such as ‘Broca’s area’ for production. tion. To fully understand human brain function is to know the causal The development of neurophysiological recording techniques in the chains of events at the molecular, cellular, population, and network twentieth century led to Barlow’s ‘ doctrine’, according to which levels that give rise to psychological function. For this , the power the functions of individual can be extrapolated to explain the to identify causal relationships is a crucial dimension of difference function of the brain as a whole. The cognitive studies of among methods. the 1980s focused on subtractive comparisons between cognitive tasks Some methods used in the study of human brain function provide meant to isolate specific cognitive operations, and led to a relatively relatively little insight into causal mechanisms. This includes methods modular view of brain function as involving localized and separable that exploit naturally occurring variation by observing the strength of regions that implement elementary mental operations. association between individual differences in brain function and beha- The methods of contemporary human neuroscience have provided a viour. Analysis of relationships between behavioural traits, , brain much more complex and nuanced view of the human brain as a dynamic structure, and brain function exemplify this approach (see Box 1 for a network with multiple levels of organization, in which function is char- discussion of genomic approaches). For many important psychological acterized by a balance of regional specialization and network integ- phenomena, from effects of life history to personality traits, we are limited ration. Although current methods are limited in their utility for to observational methods. For example, individual differences in the per- studying brain function at fine-grained levels of organization (such as sonality trait of impulsiveness have been associated with differences in single neurons or cortical columns), human neuroscience has nonethe- striatal release1, fMRI activation2, and cortical less made remarkable progress in understanding basic aspects of func- volume3. Observed associations between neural and psychological traits tional organization, and with this have come a number of applications to do not necessarily imply a causal relationship, as these associations could address real-world problems. Our goal here is to review the current state result from an unmeasured third variable that independently influences of human neuroscience, focusing on what kinds of questions can and the two measures. Nevertheless, such associations provide a valuable cannot be answered using current techniques and how those answers are starting point for theorizing about the neural mechanisms of human relevant to real-world applications. , and their evidentiary value can be strengthened by measuring possible confounds to rule them in or out. How can we study the human brain? Although , electroencelphalography/mag- Methods for studying human brain function can be organized according netoencelphalography (EEG/MEG) and single-cell recordings are some- to the kinds of mechanistic insights that each technique provides. As times criticized as being purely correlative and therefore uninformative shown in Table 1 the first characteristic is the level of mechanism cap- about mechanism, that criticism is only partly accurate. When psycho- tured by the method. Mechanisms range from the molecular level (neu- logical processes are experimentally manipulated by presenting a certain rotransmitters and receptors) to large-scale networks (the dynamic kind of stimulus and/or engaging the subject in a task, we can infer integration and coordination of different functional areas of the brain). that any reliably elicited brain activity was caused by performing these Although this distinction is related to physical scale, it does not depend psychological functions. We cannot, however, infer with confidence on the method’s spatial resolution per se. For example, positron emis- that the observed brain activity is causally responsible for the psycho- sion tomography (PET) using ligands measures logical process under study. Despite this limitation (which is shared by molecular mechanisms, even though its spatial resolution is on the order neuronal recordings in non-human animals), neuroimaging studies

1Department of Psychology, Stanford University, Stanford, California 94305, USA. 2Center for Neuroscience & Society, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA.

00 MONTH 2015 | VOL 000 | NATURE | 1 RESEARCH REVIEW

Table 1 | An overview of the levels of analysis and levels of causal inference afforded by different human neuroscience methods Level of mechanism

Molecules Cells Populations NETWORKS Strength of causal Purely observational Genetic associations with Structural morphometry N Resting functional evidence (associations do not behaviour, brain function or correlated with psychological connectivity (fMRI, necessarily imply causal brain structure traits EEG/MEG) or structural relations between mind Postmortem studies of connectivity (sMRI, DTI) and brain) expression correlated with psychological traits Correlations of MRI spectroscopy or PET ligand imaging with psychological traits Manipulate psychological Task modulation studies using Intracerebral Task activation studies N Task-based functional process and observe brain PET with neurotransmitter recording in (PET, fMRI, EEG/MEG) connectivity (fMRI, (neural measures may be ligands or MRI spectroscopy surgical patients Representational analysis EEG/MEG) epiphenomenal) (fMRI, EEG/MEG) Computational neuroimaging (fMRI, EEG/MEG) Manipulate brain and Pharmacological manipulation Direct brain Focal cortical lesions N Disconnection/white observe psychological (including hormones) stimulation in Transcranial magnetic matter lesions results (demonstrates surgical patients stimulation causal effect of neural Transcranial electrical stimulation system in behaviour) Cortical surface electrode stimulation in surgical patients

DTI, diffusion tensor imaging; EEG/MEG, /; fMRI, functional MRI; MRI, magnetic resonance imaging; PET, positron emission tomography; sMRI, structural MRI. in which psychological processes are manipulated comprise the More decisive evidence concerning causal necessity can be obtained majority of current human neuroscience research, and have advanced by manipulating the brain itself to assess the resulting effect on the our understanding of human brain function, as we will discuss in more psychological process in question. Naturally occurring or surgical detail below. lesions, which provided the basis for most of what we knew about human brain function before the advent of neuroimaging, are still of great interest because they provide insight into the causal necessity of BOX 1 specific brain regions or connections. More recently developed methods Challenges of merging of brain stimulation allow for reversible inhibition or excitation of a brain area, thereby expanding our ability to test the causal role of brain neuroimaging and regions in the mechanisms of human and action. Deep brain stimulation (DBS) provides the most precise method for targeted stimu- The substantial heritability of many psychological functions has driven lation by using surgically implanted electrodes, but is limited to situa- great interest in finding genetic underpinnings of individual tions where patients are undergoing implantation for medical . differences in neural function. Twin and family studies have Use of non-invasive brain stimulation for research purposes has grown demonstrated significant heritability for both task-related BOLD rapidly in recent decades, starting with transcranial magnetic stimu- 91 92 responses and resting-state functional connectivity in fMRI. In the lation (TMS), in which pulsed magnetic fields induce currents in the past decade, a large number of studies have also reported associations brain. Various forms of transcranial electric stimulation (TES), in which between BOLD signals and common variants in candidate genes. current is delivered using external electrodes, have also been used, of Unfortunately, this approach has generally been unsuccessful in which the most common variant is transcranial direct current stimu- identifying genetic associations that are replicated in genome-wide lation (tDCS). Unlike DBS, non-invasive brain stimulation generally association studies (GWAS). For example, a striking finding from the affects larger and more superficial areas of the brain, but researchers first well-powered GWAS of genetic variants associated with brain are seeking to improve spatial resolution with new magnetic coil shapes volume was that none of the associations previously identified through for TMS and new electrode configurations for tDCS. Focused ultra- candidate gene studies were replicated at the genome-wide level88. sound is also being explored as a means to stimulate more precisely Similarly, candidate gene associations with cognitive function (such as delimited brain regions4. Pharmacological agonists and antagonists of the association between polymorphisms in the COMT gene and particular neurotransmitter systems can be used to experimentally working ) and brain activation have generally not been manipulate the human brain at the molecular level, although with confirmed in meta-analyses, and are subject to a substantial degree of imperfect specificity5. By combining each of these manipulations of publication bias93,94. Like for many other areas of genetics, this brain function with functional brain imaging, one can leverage the cau- suggests that genome-wide approaches are the most likely to lead to reliable identification of common variants related to brain structure sal information obtained through pharmacological challenges or brain and function. However, GWAS approaches require large samples (in stimulation. For example, the causal role of activity in specific brain the tens of thousands) which are very difficult to amass for task-based regions, identified using fMRI, for a particular function has been tested fMRI studies; for that reason, GWAS-based approaches to probing the by brain stimulation, using both direct cortical stimulation (for example, 7 human brain will likely be limited to resting-state fMRI and structural ref. 6) and TMS . MRI. Other strategies, such as targeted studies investigating rare variants of large effect identified using genome sequencing or studies New capabilities of fMRI using in peripheral tissues may have greater utility for Because fMRI has become the main method for the study of human genetic studies of task-based fMRI. Task-based fMRI may also be used brain function, our review focuses on this method and new ways of using to further investigate candidate variants identified on the basis of it. In the last two decades, fMRI has developed from a newly developed GWAS studies of psychiatric disorders or population variability. technique for revealing neuronal activity to being the workhorse method of cognitive neuroscience (see the recent special issue of Neuroimage on

2|NATURE|VOL000|00MONTH2015 Nature nature15692.3d 25/9/15 11:19:18 REVIEW RESEARCH the first twenty years of fMRI8). Much has been learned about the biological mechanisms underlying level dependent (BOLD) signals9,10, but still much remains to be understood, such as the roles of specific glial and neuronal cell types in the coupling of neuronal activity to blood flow (for example, refs 11, 12). This limited physiological understanding poses problems for the interpretion of fMRI data. In particular, although fMRI signals often correlate strongly with both action potentials (‘spikes’) and local field potentials, they are Robin + largely reflective of post-synaptic processes, and in some cases they can Parrot 13 be dissociated from spiking altogether . The relative sensitivity of fMRI Chair to post-synaptic processes as opposed to spiking has been seen as a Sofa – drawback by some who view spikes as the essence of brain function, Voxels but it is worth noting that this discovery has actually rekindled interest in the analysis of local field potentials in (where these signals have long been discarded) (for example, ref. 14), and suggests that fMRI may sometimes be sensitive to subthreshold signals that would be missed by analysis of spikes only. Uncertainties in relating Standard MVPA RSA fMRI to psychological, as well as physiological, processes have also been fMRI analysis Correlation matrix Robin debated, and progress has been made on this front too. From experi- Robin Chair Parrot Sofa Chair mental approaches such as adaptation paradigms for probing represen- Parrot Sofa tations to analyses of functional connectivity, fMRI is routinely used to Is there a difference answer questions about mind–brain relationships that go far beyond in activity between groups Can we distinguish items How similar are patterns at each voxel? from each group? localization15. Here we discuss three examples of new approaches to for each item? understanding human brain function with fMRI that address questions Figure 1 | Different approaches to the analysis of fMRI data. This example of representation, computational processes and network interactions depicts data from a study in which four different stimuli were presented across the brain. (two birds and two items of furniture) and response measured for each item across nine voxels; intensity of activity is depicted from blue (negative) to red Representational analyses (positive). The standard univariate fMRI analysis approach would examine the difference at each voxel between the averages of the two categories. Multi- Early work in neuroimaging focused largely on ‘’— voxel pattern analysis (MVPA) examines the multidimensional relationship identifying regions based on the mental processes that cause them to between patterns of activity, in this case projecting the nine-dimensional space be activated. This approach has provided a large body of reliable asso- of voxel patterns (the voxel vector) into a two-dimensional space and ciations between function and structure, but has not been particularly identifying a boundary that separates items from the two classes. Representa- successful in providing new insights into how psychological functions tional similarity analysis (RSA) examines the correlations between activity are implemented16. However, two relatively recent approaches, known patterns for each item, in this case showing that items within category show a as multi-voxel pattern analysis (MVPA)17 and representational similar- high correlation (red), whereas the correlation of items between categories is ity analysis (RSA)18, can more directly relate psychological contents to low (blue). brain function (Fig. 1). MVPA involves the use of methods from the field of machine to decode or predict psychological states of stimuli (such as the similarity or typicality of objects) and the from patterns of brain activation across voxels (hence the term ‘brain- neural patterns associated with those stimuli25,26. Because psycho- reading’). Since its introduction more than a decade ago, MVPA logical theories often make predictions regarding the similarity of has been used in a number of domains to demonstrate the predictive different stimuli, RSA has also enabled the direct testing of theories, ability of fMRI activation patterns. Perhaps the most impressive are such as theories about how categories are represented27 and theories of demonstrations of the ability to successful reconstruct visual scenes19 how repeated experiences lead to enhanced learning28. RSA can be and faces20 from BOLD activity patterns; similar advances have been applied to any kind of multidimensional data, and this has enabled made for higher cognitive functions such as word meaning21. These the demonstration of systematic mappings of visual object representa- studies go beyond simply differentiating between experimental condi- tions between (using fMRI) and non-human (using tions, as they show how the underlying representational spaces relate to electrophysiological recordings)29—an example that highlights how brain activity; for example, Huth and colleagues22 developed a model human neuroscience can also help to establish more direct parallels that estimated the response at each location on the cortical surface to a with findings in non-human models, allowing insights to filter in large number of visual and semantic features present in natural movies both directions. (Fig. 2). MVPA approaches have also provided new insights into the Although much MVPA and RSA work (as depicted in Fig. 1) has neural organization of cognitive functions. For example, MVPA has focused on the representations found in localized brain regions, these informed our understanding of the mechanisms of visual , by methods are equally useful for assessing representations that are spread showing that attention changes both the representation of stimuli across the brain. For example, recent work has shown that mental states across regions of visual as well as the mutual information such as physical can be decoded by analysis of patterns of activation between regions23. In the domain of memory, MVPA has been used across brain regions30. to show that competition between memory representations in working The legitimate enthusiasm about these methods is tempered by lin- memory leads to poorer subsequent memory for those items, dem- gering questions regarding the interpretation of multivariate ana- onstrating a nonmonotonic relationship between competition and sub- lyses31,32. In addition, recent work combining electrophysiology and sequent memory24. fMRI in non-human primates has demonstrated that the sensitivity of Whereas MVPA is generally used to decode individual psycho- MVPA is limited by the spatial characteristics of the neuronal represen- logical states, RSA instead asks how the patterns of brain activity tations that code for particular features, such that some kinds of neur- evoked by different stimuli are related to one another, and thus onal patterns may be more difficult to decode using MVPA than provides the means to directly address questions of how mental repre- others33. Finally, it is important to stress that, like standard neuroima- sentations are implemented in the brain. RSA has enabled the demon- ging approaches, MVPA and RSA approaches do not inform about stration of direct isomorphisms between psychological representations causal mechanisms.

00 MONTH 2015 | VOL 000 | NATURE | 3 RESEARCH REVIEW

a are multiple RL signals in the brain, some reflecting the simple asso- S1HS1H ciation between actions and values (known as ‘model-free’ RL) and others reflecting more complex contextual and hierarchical learning processes (known as ‘model-based’ RL)37,38. Similarly, in the study of S2HS2H BrocaBroca memory, progress has been made in the mapping of medial temporal TOSTOS lobe subregions to specific computational operations such as pattern EEBABA ACAC completion and pattern separation (for example, ref. 39). In each of VV3B3B these domains, the computational interpretation of neuroimaging sig- nals has been greatly enhanced by parallel studies in non-human ani- V4V4 FBAFBA VV22 LLOO mals, allowing imaging signals to be linked more directly to direct V3V3 FFAFFA measures of neuronal activity.

PPAPPA VV11 Functional connectivity analysis and resting-state fMRI FFFAFA Perhaps the most revolutionary development to arise from human neuroimaging research is the realization that the resting brain is far b from quiescent, and that important insights into brain function can be CChangehange sselfelf MMoveove selfself AnAnimalsimals CChangehange gained from studying the correlated fluctuations of signals across the brain at rest. Much of the research into the resting state has focused on TTextext UUngulatesngulates TTalkalk asetofregions(includinganteriorandposteriormidlineregions, CCarnivoresarnivores TTravelravel SShapeshapes lateral temporoparietal cortex, and the medial , known 40 IInsectsnsects GGroupsroups as the ‘default mode’ network ) that are consistently less active during TTouchouch FFishish performance of difficult tasks41, and are functionally connected in the AtAthleteshletes resting state42. Similar patterns of resting connectivity have been BBirdsirds CColoursolours 43 44 RReptileseptiles observed in non-human primates and awake rodents , suggesting BBreathereathe MMoveove AcActivitiestivities that they reflect fundamental principles of mammalian brain - ization. There is also growing evidence that these networks may be PPlantslants PPeopleeople EvEventsents important in brain disorders. For example, the posterior portion of the DDevicesevices SSkyky appears to play a critical role in the memory deficits observed in Alzheimer’s disease, showing a convergence CContainersontainers of deposition, structural atrophy, and decreased metabolic BBodyody ppartsarts activity45. RRoadsoads Data collected in the resting state can provide insights into the FFurnitureurniture PPlaceslaces broader functional organization of the brain as well. In particular, the WWaterater organization of resting state signals bears a close relation to the organ-

CClothinglothing ization of brain activity evoked by mental tasks. For example, Smith et al.46 used independent component analysis to identify spatially inde- TToolsools OOutdoorsutdoors MMaterialsaterials pendent sets of voxels from resting-state fMRI data and from task-based RRoomsooms data (obtained from the Brainmap meta-analytic database), and demon- VVehiclesehicles BBuildingsuildings strated that the components extracted from resting-state fMRI showed a high degree of concordance with those extracted from task-based data. Figure 2 | A mapping of high-dimensional semantic space onto the cortical The overlap between resting-state and task-based functional organiza- surface. Here, voxel patterns for 1,705 different action and object categories, tion can also be seen within individuals; for example, the longitudinal 22 based on brain activity obtained during viewing of natural movies are mapped examination of a single individual revealed reliable spatial parcellation onto the cortical surface image generated using online browser at (http:// of activity in the (using resting fMRI data) that mapped gallantlab.org/semanticmovies/). a, Mapping of semantic categories to each point on the surface; the colours on the surface map to the semantic map in systematically to the activation patterns observed across a large number 47 panel b. b, A depiction of the semantic space for a specific surface point in the of task measurements . extrastriate body area (EBA). Data from ref. 22. Despite the substantial excitementaroundresting-statefMRIfind- ings, numerous concerns have been raised about their interpretation. Integrating fMRI and computational modelling In particular, there are lingering questions regarding the ways in Computational models play a central role in our understanding of both which artefacts related to head motionandphysiologicalfluctuations cognitive and brain functions and, increasingly, of the relationship may influence estimates of resting state connectivity, and whether between the two. By making assumptions explicit, computational mod- common data analytic methods may induce systematic artefacts48,49. els enable more direct testing of theories, as well as providing the means In addition, potential confounds such as light sleep50 may drive dif- to link computations at the neuronal level with higher-order functions. ferences in resting state signals. Theunconstrainednatureofresting- An example of an area in which substantial progress has been made state fMRI is a double-edged sword; it is potentially very useful for the using this approach is reinforcement learning, in which an animal study of clinical groups for whom task performance may be difficult, selects actions and learns from the rewards gained from those actions. but at the same time, it is not possible to determine whether group Computational models of reinforcement learning (RL) have long played differences reflect fundamental differences in functional connectivity a central role in artificial and psychology, and the discovery or relative differences in the ongoing mental content of different by Schultz and colleagues34 that dopamine neurons appear to signal one groups during rest (see ref. 51). of the important quantities in these models (reward prediction error) has brought these models to the forefront of the neuroscience of decision Applications of human neuroscience making. For example, a set of publications in 2003 applied RL models to With the development of new methods have come attempts to apply neuroimaging data and thereby identified correlates of reward predic- them to real-world problems, in both medical and non-medical con- tion error signals in dopaminergic target regions such as the ventral texts. (See Box 2 for a discussion of the ethical, legal, and societal issues striatum35,36. Subsequent neuroimaging work has established that there raised by these applications.)

4|NATURE|VOL000|00MONTH2015 Nature nature15692.3d 25/9/15 11:19:19 REVIEW RESEARCH

Criteria (RDoC)53—that describes disorders according to impairments BOX 2 in specific functional systems of the brain (such as fear or reward learn- Ethical, legal and societal impact of ing) and at different levels of mechanism of the kind represented in Table 1 (for example, molecules or circuits). RDoC characterizations human neuroscience cut across traditional diagnostic categories and are intended to capture the underlying pathophysiology more accurately. Given the multiple As the methods of human neuroscience find broader application, they levels of mechanism captured by the RDoC, the system encourages affect human life in new ways. The field of ‘’ is concerned research with a broad array of methods to identify potentially targetable with ethical, legal and societal issues raised by these new dysfunctions. applications95. The application of several human neuroscience methods has led to Two kinds of problems have emerged from the increasing ability of the development of targeted treatments, for example, in the field of brain imaging to reveal aspects of individual psychology: problems depression. studies have highlighted the role of that arise from the current and imminent capabilities of these the subgenual anterior in a network of regions involved methods, and problems that arise from their lack of claimed in mood, leading Mayberg and colleagues to use deep brain stimulation capabilities. To the extent that imaging can predict important personal in this area to regulate mood in depressed patients54. Lateral prefrontal characteristics such as health status, academic achievement, and regions, implicated through imaging studies in depression, have been criminal behaviour, its use must be managed with care to protect targeted with non-invasive brain stimulation, including the FDA- privacy and avoid discrimination96. To the extent that imaging cannot approved use of TMS for treatment-resistant depression. Functional provide help with high-stake problems, the public should be protected neuroimaging can itself be used as a treatment, by providing patients from claims that it can. For example, a seemingly ‘scientific’ method with a real-time measure of regional brain activity to use as a biofeed- for detecting lies or diagnosing psychiatric disorders66,97 has a strong back signal. This approach is being tested for the treatment of chronic appeal to the general public who cannot be expected to appreciate the pain, depression and addiction55. In contrast, neuroimaging has not so gap between what is claimed and what is established fact. far been very successful in aiding of disorders in New ways of changing brain function pharmaceutically, and with terms of current diagnostic categories. A recent large meta-analysis electromagnetic stimulation, also raise new ethical issues. Of course, identified a set of regions in which structural abnormalities were con- humanity has long manipulated brain function to modify mental sistently associated with psychiatric disorders, but found very little spe- states using substances such as alcohol and caffeine. However, cificity for individual disorders56, consistent with the notion that current psychopharmacology has broadly penetrated our everyday lives and diagnostic distinctions are not biologically realistic categories. the scope of psychiatric diagnoses and treatment has expanded—a societal shift that some find troubling98. Furthermore, many now use Another approach to the discovery of therapeutic targets is the use of psychoactive purely for enhancement of healthy brain function genetic association studies to identify sets of genes that are associated rather than to treat a medical condition99. Aside from issues of safety with a disorder and that together may indicate particular molecular and efficacy, brain enhancement raises issues of fairness (is it akin to pathways underlying the disorder. Although the numbers of subjects doping in sports?), justice (will the ability to access enhancements needed to establish reliable genetic associations is daunting, progress has widen the already existing gaps between haves and have-nots?) and been made through large international collaborations. For example, social standards (will unenhanced job performance become sub- Psychiatric Genomics Consortium has to date identified more than 57 standard?). 100 common genetic variants reliably implicated in . Non-invasive brain stimulation is the newest method for brain Imaging can also be used to develop endophenotypes (or intermediate enhancement. Simple transcranial electrical stimulation (tDCS)eg phenotypes) that may bear a closer relation to the effect of a gene variant devices are available to consumers at relatively low cost and regulation than does disease diagnosis, as well as to mitigate the problem of het- is minimal100. Given the public interest in this method and the erogeneity within conventional diagnostic categories (see Box 1). rudimentary state of knowledge about its effects, it is crucial that the It may be less surprising that the methods developed for human safety and efficiacy of these methods are established. The efficacy of neuroscience research have been applied in the diagnosis and treatment cognitive enhancement with tDCS is hotly debated101 and whether of neurological diseases, but at least two recent developments deserve long-term use of tDCS is safe has yet to be studied. In addition, mention here. Studies of Alzheimer’s disease at mechanistic levels from neuroethical issues of fairness, justice and social standards mentioned molecules to systems have improved diagnostic accuracy and have 58 above also apply to enhancement of brain function by brain enabled a degree of prediction before clinical signs of the disease . stimulation. Molecular biomarkers from blood and CSF, and patterns of brain activ- ity and structure have revolutionized clinical research in this area by facilitating trials of preventive treatment and by providing intermediate Brain disorders phenotypes as early gauges of effectiveness. Disorders of The methods of human neuroscience hold particular promise for under- following severe are another area of clinical neuroscience standing and treating psychiatric disorders, because these disorders do for which neuroimaging shows promise. Some patients who have been not have clear analogues in non-human animals, and animal models diagnosed as being in the vegetative state can follow commands to currently used for preclinical screening of potential therapies are imagine actions that activate specific areas of the brain in much the increasingly regarded as being inadequate52. In the absence of valid same way as healthy control subjects do, and can even use these ima- animal models, it becomes all the more crucial to apply new methods gined actions to answer questions (for example, “Do you have any for understanding human brain function and dysfunction. The goal of brothers? If yes, imagine playing tennis, if no, imagine walking through improving the treatment of neuropsychiatric disorders is made even your house.”)59. Thus, neuroimaging offers new insights into the assess- more challenging because of our current diagnostic system. Although ment of consciousness, as well as the distinct problem of prognosis, in depression, schizophrenia, and other serious psychiatric disor- severely brain-damaged patients. ders have long been considered disorders of the brain, they are still diagnosed exclusively by behavioural signs and symptoms. These dia- Predicting behaviour gnostic criteria do not seem to have clear relations to the biological The ability to predict future behaviour is of value in almost every sphere processes that would be targeted by new medical treatments. of human activity. Although it has often been said that “the best pre- In response to this problem, an alternative way of systematizing psy- dictor of future behaviour is past behaviour,” in some cases brain chiatric disorders has been developed—the NIMH Research Domain imaging can improve our ability to predict future behaviour, over and

00 MONTH 2015 | VOL 000 | NATURE | 5 RESEARCH REVIEW above what we can do with behavioural history. Marketing professionals New technologies for imaging and manipulating the were among the first to attempt to predict behaviour using brain human brain imaging. Recognizing the limitations of focus groups and other tra- Rapid advances in non-human neuroscience have been driven by the ditional methods to discern what consumers want, they have used func- development of technologies that measure and manipulate brain func- tional neuroimaging to predict the effects of different advertising tion with increasing precision. Human neuroscience has lagged in this campaigns, packaging, and other factors on consumer behaviour, based respect, in part because of the ethical challenges associated with direct on the premise that activity in the brain’s reward or motivation centres manipulation and neuronal recording of the human brain. However, in may be a more direct measurement of wanting than are verbal self- response to the urgent need for better treatments for psychiatric dis- reports60. Although most of this work is conducted by and for corpora- orders, research is underway with the aim to design implantable systems tions aiming to improve sales rather than share scientific knowledge, for sensing and modulating human brain networks68. The development published academic studies have begun to lend some credence to the of optogenetic and ‘opto-fMRI’ approaches in non-human primates69 potential of . For example, when teenage subjects were suggests that these methods may one day become feasible for use in scanned while listening to unfamiliar songs, the reward system activity human studies, and it is likely that electrical brain stimulation will be evoked by the songs, but not the subjects’ ratings of their likeability, was supplemented in the near future with optogenetic approaches. Although 61 xxxxxxxxx predictive of sales of the songs over the subsequent three years . such invasive techniques will likely only be used in rare clinical cases Prediction is also important outside of business. Falk and colleagues (that is, patients are undergoing implantation for medical reasons), have adapted neuromarketing methods for the purpose of creating they have the potential to provide much greater specificity in circuit more effective public service announcements. They showed that brain mapping. responses (but not ratings) to an anti-smoking advertisement were fMRI will probably remain the principal neuroimaging method in 62 predictive of subsequent call volume to an anti-smoking hotline . humans in the foreseeable future. However, the ongoing BRAIN initiat- 63 Gabrieli et al. recently summarized evidence concerning neuroima- ive in the United States70 is providing substantial funding to develop ging-based prediction in domains ranging from healthful eating to entirely new techniques for imaging of brain function, and a significant criminal recidivism, including numerous examples of prediction of proportion of this funding will go specifically towards the development educational outcomes. Indeed, neuroimaging can predict future aca- of new methods for imaging the human brain. In addition, new devel- demic skills over and above traditional behavioural predictors, thus opments in MRI have greatly increased the utility of standard MRI enabling earlier and more appropriate interventions to address indi- systems. For example, multiband imaging techniques71 have enabled a vidual children’s reading and math difficulties. These authors also several-fold increase in the temporal resolution of fMRI acquisitions, pointed out a number of methodological challenges in neuroima- and higher MRI field strengths (7 tesla and higher) hold promise ging-based prediction of behaviour, including the need to develop to enable improvements in spatial resolution as well (for example, and test predictions with different samples, to avoid the ‘optimism ref. 72). There is thus great reason to be optimistic that methodological bias’ that occurs when predictions are tested in the same population limits will continue to be pushed in the future. from which they were generated. Additional insight into human brain function will likely come from the study of postmortem human , which has long been a staple Human neuroscience in the courtroom method for the characterization of anatomical structure and study of In recent years the methods of human neuroscience have found their brain disorders. New techniques have enhanced the ability to visualize way into the courtroom. Perhaps the most obvious, but also the most the structure of human brain tissue (Fig. 3). For example, optical coher- misunderstood, role for neuroscience is in helping to determine criminal ence tomography has been used to image ex vivo human cortical tissue, responsibility. Proving that a criminal act may have had a neural cause is providing high-resolution imaging of cytoarchitecture with less distor- not in itself exculpatory, as every human act is caused by the brain64. tion than standard microscopy techniques73. The first whole-brain atlas However, to the extent that neuroscience can provide evidence of mental of genome-wide gene expression in postmortem human brains74 has dysfunction (for example, a tumour in the frontal cortex that may have provided an important resource for understanding how gene expression impaired the ability to control behaviour), immaturity or other psycho- relates to brain function; for example, the maps from this project have logical grounds for reduced criminal responsibility, it is potentially rel- evant and has been used. For example, the Supreme Court explicitly cited neuroscience evidence in its decision in Graham v. Florida to ab abolish life in prison without parole for juveniles who commit non- homicidal offences. It is more difficult to make legal arguments for applying neuroimaging evidence to individual cases because most find- ings from neuroimaging research are generalizations based on groups of people and may therefore not allow reliable inferences regarding indi- viduals65. Nevertheless, neuroimaging scans from defendants are some- times presented in the sentencing phase of criminal trials as grounds for mitigation of the sentence, as weaker evidentiary standards apply in the sentencing phase. Neuroimaging can be applied in ways other than determining degree of responsibility. Lie detection is one example that has been pursued in legal contexts, although it has not so far been admitted into US courts Figure 3 | New methods for characterizing the postmortem human brain. and has yet to demonstrate validity, reliability or resistance to counter- a, A map of expression of the receptor 3B displayed on the measures outside of the laboratory66. Another application concerns reconstructed cortical surface in one individual from the pain: brain-based biomarkers for pain would help discriminate real Human Brain data set (generated using data from http://human.brain- suffering from malingering—a pivotal issue in many lawsuits—and have map.org/). b, Optical coherence tomography imaging of the human brain been admitted as evidence in at least one US case67. (2.9 in-plane resolution). Large panel presents an average intensity projection in depth over 300; inset zooms are maximum intensity projections over 300, showing fibres in the (pink inset), fibres arcing through the Challenges and future directions for neuroimaging subcortical junction to insert into the cortex (cyan inset), and neurons in the The field of neuroimaging is growing rapidly, and there are a number of cortex (bright spots in the green inset). Image courtesy of Bruce Fischl, Caroline exciting new directions on the horizon. Magnain and David Boas, Massachusetts General Hospital.

6|NATURE|VOL000|00MONTH2015 Nature nature15692.3d 25/9/15 11:19:19 REVIEW RESEARCH been used to identify expression differences across different resting-state matter connectivity disruptions associated with cognitive disorders such networks75. Continued development of such resources will be essential as dyslexia77; however, diffusion imaging has inherent biases that limit for progress in understanding the genetic architecture of brain function its ability to accurately track connections across the entire brain78,79. The and their relation to mental health disorders. last decade has seen a proliferation of approaches to model functional connectivity on the basis of functional MRI data, though the dust has yet to settle regarding which methods are most effective (for example, The Human Project76 is nearing completion, and has ref. 80). To determine this, the analysis methods must be validated, already provided a rich database for the modelling of functional and which is challenging to do in humans but may be achieved using direct anatomical connectivity of the human brain. However, fundamental measurements of functional connectivity from invasive human challenges remain. For example, diffusion MRI provides the means to approaches and non-human animals to validate the neuroimaging track white matter pathways (Fig. 4) and has been used to identify white results. There is increasing evidence that at least in non-human primates

FRO FRO

INS INS

LIM IM

L TEM

TEM

PAR

PAR

OC C OCC

SBC SBC CEB BST CEB

Figure 4 | A ‘’90 for an example healthy adult female subject. The set of five rings (from the outside inward) reflect grey matter volume, area, The outermost ring shows the various brain regions arranged by lobe (fr, thickness, curvature, and connectivity density. The lines inside of the circle frontal; ins, insula; lim, limbic; tem, temporal; par, parietal; occ, occipital; nc, represent the computed degrees of connectivity between segmented brain non-cortical; bs, brain stem; CeB, ) and further ordered anterior regions using diffusion , with colour representing the relative (top) to posterior (bottom). The colour map of each region is lobe-specific and fractional anisotropy of the connection (from blue to red). Image courtesy of maps to the colour of each regional parcellation as determined using FreeSurfer. Jack Van Horn, University of Southern California.

00 MONTH 2015 | VOL 000 | NATURE | 7 RESEARCH REVIEW functional connectivity reflects anatomical connectivity as measured 6. Parvizi, J. et al. Electrical stimulation of human fusiform -selective regions 81 82 distorts face . J. Neurosci. 32, 14915–14920 (2012). using either diffusion MRI or anatomical tract-tracing ; but it remains This study uses the combination of fMRI and intracranial electrical stimulation an important challenge to establish the ways in which functional and to demonstrate the causal role of fusiform regions in face perception. diffusion connectivity measures converge or diverge. 7. Chen, A. C. et al. Causal interactions between fronto-parietal central executive and default-mode networks in humans. Proc. Natl Acad. Sci. USA 110, 19944–19949 (2013). Reproducibility of neuroimaging research 8. Bandettini, P. A. Twenty years of functional MRI: the science and the stories. Large-scale meta-analyses have made it clear that neuroimaging results Neuroimage 62, 575–588 (2012). 9. Logothetis, N. K., Pauls, J., Augath, M., Trinath, T. & Oeltermann, A. can be highly convergent across studies, to the degree that cognitive Neurophysiological investigation of the basis of the fMRI signal. Nature 412, processes can be accurately inferred from individual subject data using 150–157 (2001). decoders trained on meta-analytic data based on reported activation 10. Attwell, D. et al. Glial and neuronal control of brain blood flow. Nature 468, 83 232–243 (2010). coordinates . However, the last few years have also seen increasing con- 11. Hillman, E. M. C. Coupling mechanism and significance of the bold signal: a cern regarding the reproducibility of research findings in neuroscience, status report. Annu. Rev. Neurosci. 37, 161–181 (2014). paralleling more general concerns about reproducibility of scientific 12. Sirotin, Y. B. & Das, A. Anticipatory haemodynamic signals in not results84. These issues are particularly acute for neuroimaging given the predicted by local neuronal activity. Nature 457, 475–479 (2009). 13. Thomsen, K., Offenhauser, N. & Lauritzen, M. Principal neuron spiking: neither high dimensionality of the data, relatively low statistical power of many necessary nor sufficient for cerebral blood flow in rat cerebellum. J. Physiol. studies85, high degree of analytic flexibility in data analysis procedures86, (Lond.) 560, 181–189 (2004). and potential for questionable research practices such as circular 14. Einevoll, G. T., Kayser, C., Logothetis, N. K. & Panzeri, S. Modelling and analysis of 87 local field potentials for studying the function of cortical circuits. Nature Rev. analysis procedures . The field of neuroimaging has been at the forefront Neurosci. 14, 770–785 (2013). of a number of developments that aim to improve reproducibility and the 15. Farah, M. J. Brain images, babies, and bathwater: critiquing critiques of sharing of data are increasingly being embraced. The Alzheimer’s Disease functional neuroimaging. Hastings Cent. Rep. 44, S19–S30 (2014). 16. Poldrack, R. A. & Yarkoni, T. From brain maps to cognitive ontologies: informatics Neuroimaging Initiative (ADNI), International Neuroimaging Data and the search for mental structure. Annu. Rev. Psychol. http://dx.doi.org/ Sharing Initiative (INDI), ENIGMA, and the Human Connectome 10.1146/annurev-psych-122414-033729 (in the press). Project together have shared thousands of neuroimaging data sets and 17. Norman, K. A., Polyn, S. M., Detre, G. J. & Haxby, J. V. Beyond mind-reading: multi- voxel pattern analysis of fMRI data. Trends Cogn. Sci. 10, 424–430 (2006). this has enabled a number of novel discoveries. For example, data sharing 18. Kriegeskorte, N., Mur, M. & Bandettini, P. Representational similarity analysis— by the ENIGMA consortium has enabled the first well-powered genome- connecting the branches of . Frontiers Syst Neurosci 2, 4 wide association study of brain volume88, identifying replicated associa- (2008). 19. Naselaris, T., Prenger, R. J., Kay, K. N., Oliver, M. & Gallant, J. L. Bayesian tions between brain volume and several common genetic variants. In reconstruction of natural images from human brain activity. Neuron 63, addition, nearly all of the main software packages for neuroimaging data 902–915 (2009). analysis are free and open source, providing transparency and repro- 20. Cowen, A. S., Chun, M. M. & Kuhl, B. A. Neural portraits of perception: ducibility in data analysis across groups, and the publication of fully reconstructing face images from evoked brain activity. Neuroimage 94, 12–22 (2014). reproducible analysis workflows has begun (for example, ref. 89). The 21. Mitchell, T. M. et al. Predicting human brain activity associated with the meanings increasing use of methods, with their focus on out-of- of nouns. Science 320, 1191–1195 (2008). sample generalization rather than statistical significance, is also leading to An outstanding example of the use of fMRI along with a model of word meaning derived from a large text corpus to predict activation patterns a greater emphasis on achieving reproducibility. associated with words. 22. Huth, A. G., Nishimoto, S., Vu, A. T. & Gallant, J. L. A continuous semantic space Outlook describes the representation of thousands of object and action categories across the human brain. Neuron 76, 1210–1224 (2012). The use of new tools for imaging and manipulating the brain will con- 23. Sprague, T. C., Saproo, S. & Serences, J. T. Visual attention mitigates information tinue to advance our understanding of how the human brain gives rise to loss in small- and large-scale neural codes. Trends Cogn. Sci. 19, 215–226 thought and action. The combination of myriad methods with different (2015). 24. Lewis-Peacock, J. A. & Norman, K. A. Competition between items in working and complementary strengths and weaknesses will allow neuroscientists memory leads to forgetting. Nat. Commun. 5, 5768 (2014). to develop a multilevel understanding of the brain, spanning from mole- 25. Charest, I., Kievit, R. A., Schmitz, T. W., Deca, D. & Kriegeskorte, N. Unique cules to large-scale networks. New analysis methods have advanced semantic space in the brain of each beholder predicts perceived similarity. Proc. Natl Acad. Sci. USA 111, 14565–14570 (2014). fMRI beyond ‘blobology’ and will provide direct insight into the map- 26. Davis, T. & Poldrack, R. A. Quantifying the internal structure of categories using a ping of mental and neural representations, while newer analysis and neural typicality measure. Cereb. Cortex 24, 1720–1737 (2014). acquisition newer methods will offer other novel insights into the rela- 27. Mack, M. L., Preston, A. R. & Love, B. C. Decoding the brain’s for categorization from its neural implementation. Curr. Biol. 23, 2023–2027 tion of mind and brain. fMRI and other human neuroscience methods (2013). will continue being applied to solve real-world problems, within medi- 28. Xue, G. et al. Greater neural pattern similarity across repetitions is associated with cine and beyond. Although some of these applications are currently better memory. Science 330, 97–101 (2010). 29. Kriegeskorte, N. et al. Matching categorical object representations in inferior premature relative to the demonstrated capabilities of the methods, it temporal cortex of man and monkey. Neuron 60, 1126–1141 (2008). is clear that the new methods of human neuroscience will have much to This paper applies representational similarity analysis to human fMRI and offer science and society. monkey electrophysiology data to demonstrate similar representational spaces in the inferior temporal cortex across species. Received 31 May; accepted 4 September 2015. 30. Wager, T. D. et al. An fMRI-based neurologic signature of physical pain. N. Engl. J. Med. 368, 1388–1397 (2013). Published online XX 2015. 31. Davis, T. et al. What do differences between multi-voxel and univariate analysis mean? How subject-, voxel-, and trial-level variance impact fMRI analysis. 1. Buckholtz, J. W. et al. Dopaminergic network differences in human impulsivity. Neuroimage 97, 271–283 (2014). Science 329, 532 (2010). 32. Todd, M. T., Nystrom, L. E. & Cohen, J. D. Confounds in multivariate pattern 2. Plichta, M. M. & Scheres, A. Ventral-striatal responsiveness during reward analysis: theory and rule representation case study. Neuroimage 77, 157–165 anticipation in ADHD and its relation to trait impulsivity in the healthy (2013). population: a meta-analytic review of the fMRI literature. Neurosci. Biobehav. Rev. 33. Dubois, J., de Berker, A. O. & Tsao, D. Y. Single-unit recordings in the macaque 38, 125–134 (2014). face patch system reveal limitations of fMRI MVPA. J. Neurosci. 35, 2791–2802 3. Schilling, C. et al. Common structural correlates of trait impulsiveness and (2015). perceptual reasoning in adolescence. Hum. Brain Mapp. 34, 374–383 (2013). 34. Schultz, W. Predictive reward signal of dopamine neurons. J. Neurophysiol. 80, 4. Legon, W. et al. Transcranial focused modulates the activity of 1–27 (1998). primary somatosensory cortex in humans. Nature Neurosci. 17, 322–329 35. McClure, S. M., Berns, G. S. & Montague, P. R. Temporal prediction errors in a (2014). passive learning task activate human . Neuron 38, 339–346 (2003). 5. Chamberlain, S. R., Mu¨ ller, U., Robbins, T. W. & Sahakian, B. J. 36. O’Doherty, J. P., Dayan, P., Friston, K., Critchley, H. & Dolan, R. J. Temporal Neuropharmacological modulation of . Curr. Opin. Neurol. 19, 607–612 difference models and reward-related learning in the human brain. Neuron 38, (2006). 329–337 (2003).

8|NATURE|VOL000|00MONTH2015 Nature nature15692.3d 25/9/15 11:19:21 REVIEW RESEARCH

37. Badre, D. & Frank, M. J. Mechanisms of hierarchical reinforcement learning in 72. Yacoub, E., Harel, N. & Ugurbil, K. High-field fMRI unveils orientation columns in cortico-striatal circuits 2: evidence from fMRI. Cereb. Cortex 22, 527–536 (2012). humans. Proc. Natl Acad. Sci. USA 105, 10607–10612 (2008). 38. Daw, N. D., Gershman, S. J., Seymour, B., Dayan, P. & Dolan, R. J. Model-based 73. Magnain, C. et al. Optical coherence tomography visualizes neurons in human influences on humans’ choices and striatal prediction errors. Neuron 69, . Neurophotonics 2, 015004 (2015). 1204–1215 (2011). A demonstration of the power of optical coherence tomography to image 39. LaRocque, K. F. et al. Global similarity and pattern separation in the human neural structure in ex vivo human brain tissue. medial temporal lobe predict subsequent memory. J. Neurosci. 33, 5466–5474 74. Hawrylycz, M. J. et al. An anatomically comprehensive atlas of the adult human (2013). brain . Nature 489, 391–399 (2012). 40. Raichle, M. E. The brain’s default mode network. Annu. Rev. Neurosci. 38, 75. Richiardi, J. et al. Correlated gene expression supports synchronous activity in 433–447 (2015). brain networks. Science 348, 1241–1244 (2015). 41. Shulman, G. L. et al. Common blood flow changes across visual tasks: II. 76. Van Essen, D. C. et al. The WU-Minn : an overview. decreases in cerebral cortex. J. Cogn. Neurosci. 9, 648–663 (1997). Neuroimage 80, 62–79 (2013). 42. Greicius, M. D., Krasnow, B., Reiss, A. L. & Menon, V. Functional connectivity in the This paper presents a broad overview of the Human Connectome Project. resting brain: a network analysis of the default mode hypothesis. Proc. Natl Acad. 77. Saygin, Z. M. et al. Tracking the roots of reading ability: white matter volume and Sci. USA 100, 253–258 (2003). integrity correlate with phonological awareness in prereading and early-reading 43. Vincent, J. L. et al. Intrinsic functional architecture in the anaesthetized monkey kindergarten children. J. Neurosci. 33, 13251–13258 (2013). brain. Nature 447, 83–86 (2007). 78. Van Essen, D. C. et al. Mapping Connections in Humans and Non-human Primates: 44. Becerra, L., Pendse, G., Chang, P.-C., Bishop, J. & Borsook, D. Robust reproducible Aspirations and Challenges for Diffusion Imaging 2nd edn, Ch. 16 (, 2013). resting state networks in the awake rodent brain. PLoS ONE 6, e25701 (2011). 79. Eveley, C. et al. Superficial white matter fiber systems impede detection of long- 45. Buckner, R. L. et al. Molecular, structural, and functional characterization of range cortical connections in diffusion MR tractography. Proc. Natl Acad. Sci. USA Alzheimer’s disease: evidence for a relationship between default activity, 112, E2820–E2828 (2015). amyloid, and memory. J. Neurosci. 25, 7709–7717 (2005). 80. Smith, S. M. et al. Network modelling methods for fMRI. Neuroimage 54, This paper presents a multimodal analysis implicating the default mode 875–891 (2011). network in cognitive decline associated with Alzheimer’s disease. 81. Honey, C. J. et al. Predicting human resting-state functional connectivity from 46. Smith, S. M. et al. Correspondence of the brain’s functional architecture during structural connectivity. Proc. Natl Acad. Sci. USA 106, 2035–2040 (2009). activation and rest. Proc. Natl Acad. Sci. USA 106, 13040–13045 (2009). 82. Shen, K. et al. Information processing architecture of functionally defined This paper demonstrates that resting state networks are systematically clusters in the macaque cortex. J. Neurosci. 32, 17465–17476 (2012). associated with cognitive functions. 83. Yarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C. & Wager, T. D. Large- 47. Laumann, T. O. et al. Functional system and areal organization of a highly- scale automated synthesis of human functional neuroimaging data. Nature sampled individual human brain. Neuron 87, 657–670 (2015). Methods 8, 665–670 (2011). 48. Murphy, K., Birn, R. M. & Bandettini, P. A. Resting-state fMRI confounds and 84. Ioannidis, J. P. A. Why most published research findings are false: author’s reply cleanup. Neuroimage 80, 349–359 (2013). to Goodman and Greenland. PLoS Med. 4, e215 (2007). 49. Power, J. D., Schlaggar, B. L. & Petersen, S. E. Recent progress and outstanding 85. Button, K. S. et al. Power failure: why small sample size undermines the reliability issues in motion correction in resting state fMRI. Neuroimage 105, 536–551 of neuroscience. Nature Rev. Neurosci. 14, 365–376 (2013). (2015). 86. Carp, J. On the plurality of (methodological) worlds: estimating the analytic 50. Tagliazucchi, E. & Laufs, H. Decoding levels from typical fMRI flexibility of fMRI experiments. Frontiers Neurosci 6, 149 (2012). resting-state data reveals reliable drifts between wakefulness and . Neuron 87. Kriegeskorte, N., Simmons, W. K., Bellgowan, P. S. F. & Baker, C. I. Circular analysis 82, 695–708 (2014). in systems neuroscience: the dangers of double dipping. Nature Neurosci. 12, 51. Morcom, A. M. & Fletcher, P. C. Does the brain have a baseline? Why we should be 535–540 (2009). resisting a rest. Neuroimage 37, 1073–1082 (2007). 88. Stein, J. L. et al. Identification of common variants associated with human 52. Nestler, E. J. & Hyman, S. E. Animal models of neuropsychiatric disorders. Nature hippocampal and intracranial volumes. Nature Genet. 44, 552–561 (2012). Neurosci. 13, 1161–1169 (2010). This paper presents the first well-powered genome-wide association study of 53. Insel, T. R. The NIMH research domain criteria (RDoC) project: precision brain structure. medicine for . Am. J. Psychiatry 171, 395–397 (2014). 89. Waskom, M. L., Kumaran, D., Gordon, A. M., Rissman, J. & Wagner, A. D. 54. Mayberg, H. S. Targeted electrode-based modulation of neural circuits for Frontoparietal representations of task context support the flexible control of depression. J. Clin. Invest. 119, 717–725 (2009). goal-directed cognition. J. Neurosci. 34, 10743–10755 (2014). 55. Sulzer, J. et al. Real-time fMRI neurofeedback: progress and challenges. 90. Irimia, A., Chambers, M. C., Torgerson, C. M. & Van Horn, J. D. Circular @ Neuroimage 76, 386–399 (2013). representation of human cortical networks for subject and population-level 56. Goodkind, M. et al. Identification of a common neurobiological substrate for connectomic visualization. Neuroimage 60, 1340–1351 (2012). mental illness. JAMA Psychiatry 72, 305–315 (2015). This paper examines a large structural imaging dataset and finds that brain 91. Blokland, G. A. M. et al. Heritability of brain activation. J. abnormalities linked to mental illness are shared across diagnostic categories. Neurosci. 31, 10882–10890 (2011). 57. Schizophrenia Working Group of the Psychiatric Genomics Consortium. 92. Glahn, D. C. et al. Genetic control over the resting brain. Proc. Natl Acad. Sci. USA Biological insights from 108 schizophrenia-associated genetic loci. Nature 511, 107, 1223–1228 (2010). 421–427 (2014). 93. Barnett, J. H., Scoriels, L. & Munafo`, M. R. Meta-analysis of the cognitive effects of 58. Sperling, R. A., Karlawish, J. & Johnson, K. A. Preclinical Alzheimer disease-the the catechol-O-methyltransferase gene Val158/108Met polymorphism. Biol. challenges ahead. Nature Rev. Neurol. 9, 54–58 (2013). Psychiatry 64, 137–144 (2008). 59. Owen, A. M. Detecting consciousness: a unique role for neuroimaging. Annu. Rev. 94. Nickl-Jockschat, T., Janouschek, H., Eickhoff, S. B. & Eickhoff, C. R. Lack of meta- Psychol. 64, 109–133 (2013). analytic evidence for an impact of COMT Val158Met genotype on brain 60. Ariely, D. & Berns, G. S. Neuromarketing: the hope and hype of neuroimaging in activation during working memory tasks. Biol. Psychiatry, (2015). business. Nature Rev. Neurosci. 11, 284–292 (2010). 95. Farah, M. J. Neuroethics: the ethical, legal, and societal impact of neuroscience. 61. Berns, G. S. & Moore, S. A neural predictor of cultural popularity. J. Consum. Annu. Rev. Psychol. 63, 571–591 (2012). Psychol. 22, 154–160 (2012). 96. Illes, J. & Racine, E. Imaging or imagining? A neuroethics challenge informed by 62. Falk, E. B., Berkman, E. T. & Lieberman, M. D. From neural responses to genetics. Am. J. Bioeth. 5, 5–18 (2005). population behavior: neural focus group predicts population-level media effects. 97. Farah, M. J. & Gillihan, S. J. The puzzle of neuroimaging and psychiatric Psychol. Sci. 23, 439–445 (2012). diagnosis: technology and nosology in an evolving discipline. AJOB Neurosci. 3, 63. Gabrieli, J. D. E., Ghosh, S. S. & Whitfield-Gabrieli, S. Prediction as a humanitarian 31–41 (2012). and pragmatic contribution from human cognitive neuroscience. Neuron 85, 98. Conrad, P. The Medicalization of Society: On the Transformation of Human 11–26 (2015). Conditions into Treatable Disorders (Johns Hopkins Univ. Press, 2007). 64. Morse, S. J. Brain overclaim syndrome and criminal responsibility: A diagnostic 99. Sahakian, B. & Morein-Zamir, S. Professor’s little helper. Nature 450, 1157–1159 note. Ohio State J. Criminal Law 3, 397–412 (2006). (2007). 65. Jones, O. D., Wagner, A. D., Faigman, D. L. & Raichle, M. E. Neuroscientists in court. 100. Fitz, N. S. & Reiner, P. B. The challenge of crafting policy for do-it-yourself brain Nature Rev. Neurosci. 14, 730–736 (2013). stimulation. J. Med. Ethics 41, 410–412 (2015). 66. Farah, M. J., Hutchinson, J. B., Phelps, E. A. & Wagner, A. D. Functional MRI-based 101. Horvath, J. C., Forte, J. D. & Carter, O. Quantitative review finds no evidence of lie detection: scientific and societal challenges. Nature Rev. Neurosci. 15, cognitive effects in healthy populations from single-session transcranial direct 123–131 (2014). current stimulation (tDCS). Brain Stimul. 8, 535–550 (2015). 67. Reardon, S. Neuroscience in court: the painful truth. Nature 518, 474–476 Acknowledgements Thanks to I. Eisenberg, D. Glahn, R. Raizada, and M. Shine for (2015). comments on an earlier draft of this manuscript, and N. Logothetis for helpful 68. Underwood, E. DARPA aims to rebuild brains. Science 342, 1029–1030 (2013). discussions. 69. Gerits, A. et al. Optogenetically induced behavioral and functional network changes in primates. Curr. Biol. 22, 1722–1726 (2012). Author Contributions R.P. and M.F. planned and wrote the paper. 70. Insel, T. R., Landis, S. C. & Collins, F. S. The NIH Brain Initiative. Science 340, 687–688 (2013). Author Information Reprints and permissions information is available at 71. Moeller, S. et al. Multiband multislice GE-EPI at 7 tesla, with 16-fold acceleration www.nature.com/reprints. The authors declare no competing financial interests. using partial parallel imaging with application to high spatial and temporal Readers are welcome to comment on the online version of the paper. Correspondence whole-brain fMRI. Magn. Reson. Med. 63, 1144–1153 (2010). and requests for materials should be addressed to R.P. ([email protected]).

00 MONTH 2015 | VOL 000 | NATURE | 9 RESEARCH REVIEW

Author Queries Journal: Nature Paper: nature15692 Title: Progress and challenges in probing the human brain

Query Query Reference

1 AUTHOR: A PDF proof will be produced on the basis of your corrections to this preproof and will contain the main-text figures edited by us and the Extended Data items supplied by you (which may have been resized but will not have been edited otherwise by us).

2 When you receive the PDF proof, please check that the display items are as follows (doi:10.1038/nat- ure15692): Figs 0 (black & white); 1–3 (colour); Tables: 1; Boxes: 2; Extended Data display items: 0.

3 Please check the edits to all main-text figures (and tables, if any) very carefully, and ensure that any error bars in the figures are defined in the figure legends. If you wish to revise the Extended Data items for consistency with main-text figures and tables, please copy the style shown in the PDF proof (such as italicising variables and gene symbols, and using initial capitals for labels) and return the revised Extended Data items to us along with your proof corrections.

4 N A single sentence summarizing your paper (websum), which will appear online on the table of contents and in e-alerts, has been provided below. Please check this sentence for accuracy and appropriate emphasis.

5 Author: first sentence revised slightly. OK?

6 Author: this reference is not cited in the text, please cite or remove

Web This Review evaluates current techniques used to investigate human brain function, discusses the suc- summary cesses and limitations of these techniques to test hypotheses about causal mechanisms, and looks to future directions and implementation of these techniques in real-world problems.

For Nature office use only:

Layout % Figures/Tables/Boxes % References % DOI % Error bars % Supp info % Title % Colour % Acknowledgements % Authors % Text % Author contribs % Addresses % Methods % COI % First para % Received/Accepted % Correspondence % AOP % Author corrx % Extended Data % Web summary % Accession codes link %

10 | NATURE | VOL 000 | 00 MONTH 2015 Nature nature15692.3d 25/9/15 11:19:23